Escolar Documentos
Profissional Documentos
Cultura Documentos
118
Abstract
A direct design procedure is established for linear time-invariant controllable multivariable systems which
permits a straightforward calculation of the feedback matrix to attain prescribed closed-loop poles, it permits
the choice of the relative tightness of the feedback to each input, and hence allows design with feedback to
only some inputs (i.e. incomplete input feedback). Furthermore, it permits the design of the important
practical case where some states are not accessible. For these, the procedure produces the necessary and
sufficient condition for pole positioning to be possible, and permits the required feedback to be calculated.
The procedure is applicable to systems with any number of inputs and states. Computationally, the method
is extremely fast, with computation times of the order of seconds, making it well suited to computer-aided
control-system design.
Introduction
The theory of multivariable control systems is well
advanced, and several methods exist for choosing a feedback
law to achieve desired design objectives. In recent years, there
has been a considerable interest in choosing feedback laws for
controllable multivariable systems to achieve arbitrary
dynamics of the closed-loop system. Most of these techniques
make use of feedback from the state vector to assign the
closed-loop system poles to particular locations;1"4 others use
output feedback for the same purpose.5'6 The present paper
attempts to contribute to the design of state-feedback
controllers by extending a single-input design procedure to
the multi-input case.
The method proposed is based on the equivalence of the
closed-loop characteristic polynomials of a multi-input and a
corresponding single-input system. The latter is first designed
using a previously established direct design method for singleinput systems,7 and the result is then transferred back to the
former.
The method has a number of attractive features. It is
computationally very fast, with computation times of the
order of seconds, and is well suited to the computer-aided
design of control systems. It provides the designer with
complete freedom over the relative tightness of the feedback
to each input, and hence also allows design with feedback to
only some inputs, i.e. incomplete input feedback. A further
important feature is that it permits design with incomplete
state feedback, when some of the states are not accessible.
Since the method involves the use of the existing procedure
for single-input systems, this is first summarised.
1
by
40 C&S12
(3)
x
(4)
(1)
Paper 6432 C, first received 8th January and in revised form 17th
March 1971
Dr. Fallside and Mr, Seraji are with the University Engineering
Department, University of Cambridge, Trumpington Street, Cambridge
CB2 1PZ, England
i = ! , . . . , , is the
X(s) = Gc(s)V(s)
3.1
x = Ax + bu
= {gj(s)}jF(s),
x = Ax + Bu
(5)
797.
(6)
\sl - A + bp\ = 0
(7)
where b = Bq is an (n X 1) matrix.
Comparison of eqns. 6 and 7 yields the following lemma.
Lemma: The closed-loop poles of a multi-input system which
has a plant matrix A, a control matrix B and a feedback
matrix P are coincident with those of an equivalent singleinput system which has the same plant matrix A, a control
matrix b and feedback vector/;, where b = Bq and P = qp.
Making use of this equivalence, the design problem can be
solved in three straightforward steps:
(i) Choose an m-vector q. In general, q is arbitrary except
for the special cases treated in Sections 3.3, 4 and 5.
(ii) Find the /z-feedback vector p required to position the
poles of the equivalent single-input system (A, Bq) at the
desired positions A,, . . ., An, using the single-input
direct-design procedure based on eqn. 4.
(iii) For the multi-input system (A, B) the required state
feedback matrix is P = qp.
3.3
Controllability
Example 2
Example 1
q =
by
ro n
ri
01
P'l
s(s + 5)
3q2>-
F(s)
'
. . . .
'
(9)
= (s + a)(s
i =
thus,
(b - 5)
2q2) \_q2
J
\2qx\q2
~,
2qx\q2 - 2q\\q?[\
^_x\q
,2 J
2~,~ 2q
2q2_
2qx\q2~\
= qp= \
s(s + 5)
(10)
foi + 2<h)
we obtain
2 -
+ b)-
5q2]
J, 2
-ti
(8)
(11)
we obtain
P
a
= qp = 7 -
"1
Q\P\--<l\Pn
P=qp =
[p\ / > , / ] =
Slml
5.1
UlmPl
QmPn.
. . . .
(12)
4.1
1 1 -2
0 0"
01
.1 -2 0
2 0 -2 x + 0
12
1.
(13)
q=
Then, for the equivalent single-input system,
5s2- Us + 14'
4s- 26
r -
+ 5s
~P\
Pi
-P3.
0 1 0" -l
1 5 1
4 0 2.
6"
a2- 7
3 - 2.
"i -
or
10a, - 2a2 + a3 - 48 '
1-23
097]
(15)
and
3-1
6-15 4-35
(16)
1-23 0-97
0
0
0
which has the required properties.
This calculation can be carried out using the computer
program mentioned in Section 3.2.
P = qp = 0-62
1
+ 6s2 + 7.y + 2
(.17)
3.s2 + 2 ^ + 1 4
62
system
"0"
1
0
and
Example 4
2a, - 12
- 2 0 a , + 4a2 - a3 + 94
09)
UlJ
is thus det L # 0, which is always true if the system is completely controllable, and then, from eqn. 21,
pT = M(a - d)
(22)
G(s)q = (si -
+ g
j3 + 9 j 2
33
(30, - 802)
(qx - 2q2)s
qxs2 + (60, - 2q2)s + (90, - 202;
L Qis2 + (20, + 2q2)s + (60, - 502) .
where M = L~l.
For incomplete state feedback, with r states not fed back,
r elements of pT must be set to zero and eqn. 22 then implies
(23)
a(a - d) = 0
aa = p
(24)
. . . .
This imposes r conditions on the coefficients ax, . . ., an of
H(js) to be satisfied, and these, in conjunction with the
controllability requirement det L =j 0, constitute the necessary
and sufficient conditions for pole assignment with incomplete
state feedback. When these conditions hold, p can be calculated from eqn. 22.
Since the elements of a are related to the closed-loop poles,
eqn. 24 can also be written in terms of pole positions. In this
form, eqn. 24 implies certain forbidden regions for closedloop poles in the splane.
It can be noted that eqn. 24 can be interpreted as allowing
only /; r of the elements of a to be chosen arbitrarily, and
the remaining r elements of a are then given by eqn. 24. This
agrees with the complete freedom of choice of a for complete
state feedback when r = 0.
The condition of eqn. 24 can be readily computerised to
determine whether pole positioning is possible, and the feedback p can be calculated from eqn. 22. When a desired pole
pattern is unattainable, an attainable near set can be calculated by iterating on a to satisfy eqn. 24.
5.2
L(q)pT = a -
(25)
(26)
Example 5
x =
1
2
8
1
-3
2
-2'
-2
-7
x+
"0 0"
1 0
.0 1
(27)
Find the state-feedback matrix P which positions the closedloop poles at 3, 3, 3 with no feedback from xx.
We take q = (qx q2)T, a r | d the equivalent single-input
800
P2\ =
6q2^0
0"
-2
-6
. (29)
1
q2 - 4qxq2 + 6qj
L/>3J
(30)
. . .
(31)
P0
(32)
-01/0 J
\q2
using eqn. 30. Then, inserting the condition of eqn. 31, there
results
1 TO
'-9L0
-2
-41
(33>
1 2J
5.3.1
42
The existence of a feedback matrix P for pole positioning of a multivariable system (A, B) with incomplete state
feedback reduces to the existence of the feedback vector p
for the equivalent single-input system (A, Bq) with incomplete
state feedback. The analysis is thus similar in form to that of
Section 5.1.
However, now the elements of q appear in G(s)q, and hence
eqn. 21 becomes
(28)
Example 6
Consider a system described by
1 -1
0 -2
0
0
0
0
0
0
-3
0
0
0
4
x+
"0 0
1
1 0 -1
0
0 1
2
1
(34)
. . . .
i m.
i -,w
i ^/
i ^
8
1
3)}
5
= as2 + j3*2 + ys + 8
2tfi
_6^r2 -
6tf,
Uq2
P3
^ 2 )J
(37)
-PA-
Assuming that
+ 3q\ ^ 0
-Aq}
. . . .
(38)
which are the conditions for complete controllability, i.e.
det L ^ O , and setting/^ = 0 gives the condition
q_x
?2
Pi
^y2)
72 -
cc
~ P\
References
ANDERSON, B. D. o., and LUENBERGER, D. G.\ 'Design of multivariable
Acknowledgment
The authors wish to thank R. V. Patel of the University of Cambridge Computer Laboratory for collaboration in
developing the computer program referred to in Section 5.1.
qx(s
q2(s
1 6 4 a - 25ff - y + 55
167a - 19j3 - 3y + S ' * '
'
'
Appendix
and
His) - F(s) =
pigi(s)
(41)
Pi
PT>
0)
"(-49a + 5 + y)r
1
10 -
2A-
-21)
49a + 5p + y
0
j8(5 - 2r) - y
P(S-2r)-y
- 2r) + y}
r{a(31 - 16r) -
0"
a(31
(40)
0J
q' = pq. Note that this does not affect the ratios of the elements of q. In this case,
1
His) - Fis) =
Conclusions
W
(42)
/=!
Comparing eqns. 41 and 42, we concludep'i = - ph i = 1,...,
1
P
n, and hence p' = -p. Finally, for this choice of q,
P
P' = q'p' = (pq) (-pj = qp
(43)
which is unchanged. P is therefore unique for a particular
choice of ratios of the elements of q.
801